Accurate understanding of the biological functions of enzymes is vital for various tasks in both pathologies and industrial biotechnology. However, the existing methods are usually not fast enough and lack explanations on the prediction results, which severely limits their real-world applications. Following our previous work, DEEPre, we propose a new interpretable and fast version (ifDEEPre) by designing novel self-guided attention and incorporating biological knowledge learned via large protein language models to accurately predict the commission numbers of enzymes and confirm their functions. Novel self-guided attention is designed to optimize the unique contributions of representations, automatically detecting key protein motifs to provide meaningful interpretations. Representations learned from raw protein sequences are strictly screened to improve the running speed of the framework, 50 times faster than DEEPre while requiring 12.89 times smaller storage space. Large language modules are incorporated to learn physical properties from hundreds of millions of proteins, extending biological knowledge of the whole network. Extensive experiments indicate that ifDEEPre outperforms all the current methods, achieving more than 14.22% larger F1-score on the NEW dataset. Furthermore, the trained ifDEEPre models accurately capture multi-level protein biological patterns and infer evolutionary trends of enzymes by taking only raw sequences without label information. Meanwhile, ifDEEPre predicts the evolutionary relationships between different yeast sub-species, which are highly consistent with the ground truth. Case studies indicate that ifDEEPre can detect key amino acid motifs, which have important implications for designing novel enzymes. A web server running ifDEEPre is available at https://proj.cse.cuhk.edu.hk/aihlab/ifdeepre/ to provide convenient services to the public. Meanwhile, ifDEEPre is freely available on GitHub at https://github.com/ml4bio/ifDEEPre/.

Bacteria within the gut microbiota possess the ability to metabolize a wide array of human drugs, foods, and toxins, but the responsible enzymes for these chemical events remain largely uncharacterized due to the time-consuming nature of current experimental approaches. Attempts have been made in the past to computationally predict which bacterial species and enzymes are responsible for chemical transformations in the gut environment, but with low accuracy due to minimal chemical representation and sequence similarity search schemes. Here, we present an in silico approach that employs chemical and protein Similarity algorithms that Identify MicrobioMe Enzymatic Reactions (SIMMER). We show that SIMMER accurately predicts the responsible species and enzymes for a queried reaction, unlike previous methods. We demonstrate SIMMER use cases in the context of drug metabolism by predicting previously uncharacterized enzymes for 88 drug transformations known to occur in the human gut. We validate these predictions on external datasets and provide an in vitro validation of SIMMER\'s predictions for metabolism of methotrexate, an anti-arthritic drug. After demonstrating its utility and accuracy, we made SIMMER available as both a command-line and web tool, with flexible input and output options for determining chemical transformations within the human gut. We present SIMMER as a computational addition to the microbiome researcher\'s toolbox, enabling them to make informed hypotheses before embarking on the lengthy laboratory experiments required to characterize novel bacterial enzymes that can alter human ingested compounds.

Synthetic biology and metabolic engineering rely on computational search tools for predictions of novel biosynthetic pathways to industrially important compounds, many of which are derived from aromatic amino acids. Pathway search tools vary in their scope of covered reactions and compounds, as well as in metrics for ranking and evaluation. In this work, we present a new computational resource called ARBRE: Aromatic compounds RetroBiosynthesis Repository and Explorer. It consists of a comprehensive biochemical reaction network centered around aromatic amino acid biosynthesis and a computational toolbox for navigating this network. ARBRE encompasses over 33\'000 known and 390\'000 novel reactions predicted with generalized enzymatic reactions rules and over 74\'000 compounds, of which 19\'000 are known to biochemical databases and 55\'000 only to PubChem. Over 1\'000 molecules that were solely part of the PubChem database before and were previously impossible to integrate into a biochemical network are included into the ARBRE reaction network by assigning enzymatic reactions. ARBRE can be applied for pathway search, enzyme annotation, pathway ranking, visualization, and network expansion around known biochemical pathways and products of lignin degradation to predict valuable compound derivations. In line with the standards of open science, we have made the toolbox freely available to the scientific community on git (https://github.com/EPFL-LCSB/ARBRE) and we provide the web-version at http://lcsb-databases.epfl.ch/arbre/. We envision that ARBRE will provide the community with a new computational resource and comprehensive search tool to predict and rank pathways towards industrially important aromatic compounds.

The ATLAS of Biochemistry is a repository of both known and novel predicted biochemical reactions between biological compounds listed in the Kyoto Encyclopedia of Genes and Genomes (KEGG). ATLAS was originally compiled based on KEGG 2015, though the number of KEGG reactions has increased by almost 20 percent since then. Here, we present an updated version of ATLAS created from KEGG 2018 using an increased set of generalized reaction rules. Furthermore, we improved the accuracy of the enzymes that are predicted for catalyzing novel reactions. ATLAS now contains ∼150 000 reactions, out of which 96% are novel. In this report, we present detailed statistics on the updated ATLAS and highlight the improvements with regard to the previous version. Most importantly, 107 reactions predicted in the original ATLAS are now known to KEGG, which validates the predictive power of our approach. The updated ATLAS is available at https://lcsb-databases.epfl.ch/atlas.

BACKGROUND: Genome-scale metabolic modeling is a cornerstone of systems biology analysis of microbial organisms and communities, yet these genome-scale modeling efforts are invariably based on incomplete functional annotations. Annotated genomes typically contain 30-50% of genes without functional annotation, severely limiting our knowledge of the \"parts lists\" that the organisms have at their disposal. These incomplete annotations may be sufficient to derive a model of a core set of well-studied metabolic pathways that support growth in pure culture. However, pathways important for growth on unusual metabolites exchanged in complex microbial communities are often less understood, resulting in missing functional annotations in newly sequenced genomes.
RESULTS: Here, we present results on a comprehensive reannotation of 27 bacterial reference genomes, focusing on enzymes with EC numbers annotated by KEGG, RAST, EFICAz, and the BRENDA enzyme database, and on membrane transport annotations by TransportDB, KEGG and RAST. Our analysis shows that annotation using multiple tools can result in a drastically larger metabolic network reconstruction, adding on average 40% more EC numbers, 3-8 times more substrate-specific transporters, and 37% more metabolic genes. These results are even more pronounced for bacterial species that are phylogenetically distant from well-studied model organisms such as E. coli.
CONCLUSIONS: Metabolic annotations are often incomplete and inconsistent. Combining multiple functional annotation tools can greatly improve genome coverage and metabolic network size, especially for non-model organisms and non-core pathways.